Question 80
An audit preparation manager is using Claude to assess complex exceptions across several policy constraints with the policy manual and exception records. The review identifies 2 independent defects: 1) shared task context belongs across several chats but is scattered through one long thread; 2) a recurring content task requires balanced response quality rather than extreme latency or deep investigation. Which TWO actions directly address these separately identified failures? Choose TWO.
- For a deliverable reaching internal auditors, trial an appropriate Haiku model for the routine workload and check its output quality.
- For a deliverable reaching internal auditors, evaluate an Opus model on the difficult examples before adopting it.
- For a deliverable reaching internal auditors, start with a current Sonnet model and benchmark it on representative work.
- For a deliverable reaching internal auditors, create a dedicated Claude Project with reusable instructions and knowledge files.
- For a deliverable reaching internal auditors, compare candidate models on representative tasks with measured quality, delay, and cost.
Explanation
Correct Answers
C:When an audit preparation manager must use projects for recurring shared context in the assignment to assess complex exceptions across several policy constraints, the important distinction is that sonnet offers a general balance of capability and responsiveness for routine professional analysis and drafting. In practical terms for internal auditors, the issue is specific: a recurring content task requires balanced response quality rather than extreme latency or deep investigation.
D:When an audit preparation manager must use projects for recurring shared context in the assignment to assess complex exceptions across several policy constraints, here, projects give related chats persistent task context through their configured instructions and knowledge base. The issue is specific: shared task context belongs across several chats but is scattered through one long thread. This addresses the described failure in assess complex exceptions across several policy constraints; a Project does not guarantee that ad hoc content from one chat appears in every other chat.
Incorrect Answers
A:When an audit preparation manager must use projects for recurring shared context in the assignment to assess complex exceptions across several policy constraints, this choice can help a team choose haiku for efficient routine tasks: haiku emphasizes speed and efficiency, so it can fit extraction or simple classification when test results meet requirements. It does not complete the control needed to use projects for recurring shared context and to choose sonnet for balanced work.
B:When an audit preparation manager must use projects for recurring shared context in the assignment to assess complex exceptions across several policy constraints, it can be reasonable to choose opus for demanding reasoning: opus targets complex reasoning and demanding knowledge work where additional capability may justify a heavier tier. The observed failure instead requires the team to use projects for recurring shared context and to choose sonnet for balanced work.
E:A team may benchmark candidate model choices in a different situation: task-specific evaluation makes selection empirical and avoids assuming the most capable model is always the best economic choice. Here the immediate requirement is to use projects for recurring shared context and to choose sonnet for balanced work.
Question 81
A philanthropic program manager needs to verify a grant-performance narrative using milestone records and budget reports. The reported problem is this: several apparent confirmations all trace back to one unverified press release. What adjustment is most appropriate before the output goes to external donors?
- For a deliverable reaching external donors, review language, examples, and actionable detail with a representative target reader.
- For a deliverable reaching external donors, confirm the critical assertion with an independent credible source where available.
- For a deliverable reaching external donors, align units, periods, and denominators before comparing or aggregating the values.
- For a deliverable reaching external donors, test the output against the stated inclusions and exclusions, including edge cases.
Explanation
Correct Answer
B:When a philanthropic program manager must seek independent corroboration in the assignment to verify a grant-performance narrative, here, independent corroboration reduces reliance on a single potentially mistaken or misleading record. The issue is specific: several apparent confirmations all trace back to one unverified press release. This addresses the described failure in verify a grant-performance narrative; repeating the same claim across derivative sites is not genuine independent confirmation.
Incorrect Answers
A:For external donors, the team might test audience suitability: audience review measures whether the answer communicates usable decisions to the people who will act on it. When a philanthropic program manager must seek independent corroboration in the assignment to verify a grant-performance narrative, that is insufficient for this case, which requires the team to seek independent corroboration.
C:When a philanthropic program manager must seek independent corroboration in the assignment to verify a grant-performance narrative, a team may normalize units and periods in a different situation: unit and time normalization prevents apples-to-oranges conclusions, even where every source figure is individually correct. In practical terms for external donors, here the immediate requirement is to seek independent corroboration.
D:When a philanthropic program manager must seek independent corroboration in the assignment to verify a grant-performance narrative, this may help a team inspect scope exclusions: scope checks establish whether the conclusions apply to exactly the group the stakeholder requested. In practical terms for external donors, but the case instead requires a step to seek independent corroboration.
Question 82
While preparing to develop a family workshop outline, a museum education lead is working with exhibit descriptions and accessibility guidance. Claude handles common cases but guesses the formatting of rare exceptions inconsistently. Which action best targets the actual failure in this assignment?
- For a deliverable reaching volunteer facilitators, delimit quoted source content as data and state that directives inside it are not to be followed.
- For a deliverable reaching volunteer facilitators, ask Claude to identify the key missing requirements and seek clarification before drafting.
- For a deliverable reaching volunteer facilitators, provide two or three diverse examples of acceptable input-to-output mappings.
- For a deliverable reaching volunteer facilitators, tell Claude the decision to support and the specific question the deliverable must answer.
Explanation
Correct Answer
C:Because the result will reach volunteer facilitators, examples make borderline categorization and presentation choices concrete, including cases a generic rule may not cover. The issue is specific: claude handles common cases but guesses the formatting of rare exceptions inconsistently.
Incorrect Answers
A:When a museum education lead must supply representative examples in the assignment to develop a family workshop outline, a team may separate instructions from source text in a different situation: clear boundaries distinguish the user's authorized task from untrusted words that happen to appear in retrieved material. In practical terms for volunteer facilitators, here the immediate requirement is to supply representative examples.
B:When a museum education lead must supply representative examples in the assignment to develop a family workshop outline, this may help a team clarify missing requirements: resolving ambiguity at intake prevents the model from silently choosing an unapproved assumption. In practical terms for volunteer facilitators, but the case instead requires a step to supply representative examples.
D:It can be reasonable to define the intended business decision: a task framed around a decision gives the model a relevance test: facts that do not inform that decision can be left out. When a museum education lead must supply representative examples in the assignment to develop a family workshop outline, the observed failure instead requires the team to supply representative examples.
Question 83
An environmental outreach analyst wants dependable results while using Claude to review a waste-reduction dashboard narrative. The team wants to skip an authorized review step because the prose sounds expert. Which approach best aligns with the described evidence and responsibility?
- To review a waste-reduction dashboard narrative, recompute the derived number from the original inputs and check the stated denominator.
- To review a waste-reduction dashboard narrative, ask for uncertainty and missing-evidence flags, then verify decision-critical claims independently.
- To review a waste-reduction dashboard narrative, route the draft to an authorized subject-matter professional before final action.
- To review a waste-reduction dashboard narrative, confirm the critical assertion with an independent credible source where available.
Explanation
Correct Answer
C:When an environmental outreach analyst must obtain domain-expert signoff in the assignment to review a waste-reduction dashboard narrative, this is the relevant control because expert review tests the specialized judgment and accountability that language-model fluency cannot supply. The issue is specific: the team wants to skip an authorized review step because the prose sounds expert. In an environmental outreach analyst's assignment to review a waste-reduction dashboard narrative, internal formatting checks alone cannot authorize a clinical, legal, or compliance decision.
Incorrect Answers
A:When an environmental outreach analyst must obtain domain-expert signoff in the assignment to review a waste-reduction dashboard narrative, choosing to recalculate derived figures has a purpose: independent arithmetic catches denominator swaps, rounding errors, and accidental aggregation mistakes in generated summaries. In practical terms for civic stakeholders, it does not repair the failure that requires the team to obtain domain-expert signoff.
B:When an environmental outreach analyst must obtain domain-expert signoff in the assignment to review a waste-reduction dashboard narrative, a team may calibrate uncertainty in a different situation: transparent uncertainty shows reviewers where evidence is thin and where additional checking is necessary. In practical terms for civic stakeholders, here the immediate requirement is to obtain domain-expert signoff.
D:When an environmental outreach analyst must obtain domain-expert signoff in the assignment to review a waste-reduction dashboard narrative, this choice can help a team seek independent corroboration: independent corroboration reduces reliance on a single potentially mistaken or misleading record. It does not complete the control needed to obtain domain-expert signoff.
Question 84
A property-management assistant is using Claude to summarize maintenance requests with tenant messages and work-order status. The review identifies 2 independent defects: 1) the summary makes assumptions about missing dates without labeling them; 2) multiple attempts use different labeling conventions despite repeated verbal descriptions of the desired style. Which TWO actions directly address these separately identified failures? Choose TWO.
- To summarize maintenance requests, specify headings, field names, order, and expected output shape before asking for the draft.
- To summarize maintenance requests, provide two or three diverse examples of acceptable input-to-output mappings.
- To summarize maintenance requests, request separate sections for verified facts, assumptions, and questions that remain unanswered.
- To summarize maintenance requests, ask Claude to identify the key missing requirements and seek clarification before drafting.
- To summarize maintenance requests, list weighted evaluation criteria and request evidence for each option against those criteria.
Explanation
Correct Answers
B:Because the result will reach building coordinators, examples make borderline categorization and presentation choices concrete, including cases a generic rule may not cover. When a property-management assistant must distinguish facts from assumptions in the assignment to summarize maintenance requests, the issue is specific: multiple attempts use different labeling conventions despite repeated verbal descriptions of the desired style.
C:For the given inputs (tenant messages and work-order status), separating evidence from inference gives reviewers a practical place to challenge uncertain statements before use. The issue is specific: the summary makes assumptions about missing dates without labeling them. The correction still has a boundary: an assumption list makes uncertainty visible; it does not convert the assumptions into verified truths.
Incorrect Answers
A:When a property-management assistant must distinguish facts from assumptions in the assignment to summarize maintenance requests, this may help a team request an explicit response structure: a visible output specification lets Claude organize information in the exact sections needed by the receiving workflow. In practical terms for building coordinators, but the case instead requires a step to distinguish facts from assumptions and to supply representative examples.
D:When a property-management assistant must distinguish facts from assumptions in the assignment to summarize maintenance requests, for building coordinators, the team might clarify missing requirements: resolving ambiguity at intake prevents the model from silently choosing an unapproved assumption. In practical terms for building coordinators, that is insufficient for this case, which requires the team to distinguish facts from assumptions and to supply representative examples.
E:When a property-management assistant must distinguish facts from assumptions in the assignment to summarize maintenance requests, choosing to define comparison criteria has a purpose: a shared rubric makes comparisons traceable to the stakeholder's real trade-offs rather than model-generated preferences. In practical terms for building coordinators, it does not repair the failure that requires the team to distinguish facts from assumptions and to supply representative examples.
Question 85
An internal communications manager is using Claude to improve the quarterly employee-newsletter process from verified announcements and release rules. An operator is treating the model's summary as an approval rather than a draft for review. The result will go to staff. Which intervention addresses the cause rather than merely improving presentation?
- To improve the quarterly employee-newsletter process, place a named human approver between Claude's recommendation and the final decision.
- To improve the quarterly employee-newsletter process, triage tasks by complexity and risk, using lightweight handling for routine cases and review or escalation for difficult ones.
- To improve the quarterly employee-newsletter process, establish the authoritative input repositories and freshness rules for the workflow.
- To improve the quarterly employee-newsletter process, set measurable acceptance criteria and test them on representative examples.
Explanation
Correct Answer
A:For the given inputs (verified announcements and release rules), human approval preserves accountability where policy, professional judgment, or irreversible consequences are involved. The issue is specific: an operator is treating the model's summary as an approval rather than a draft for review.
Incorrect Answers
B:For staff, the team might route work by complexity: routing preserves effort for cases that need deeper analysis while reducing unnecessary work on well-defined tasks. When an internal communications manager must keep a human decision gate in the assignment to improve the quarterly employee-newsletter process, that is insufficient for this case, which requires the team to keep a human decision gate.
C:Choosing to specify trusted input sources has a purpose: approved source boundaries reduce conflicting instructions and make data lineage clearer during review. It does not repair the failure that requires the team to keep a human decision gate.
D:When an internal communications manager must keep a human decision gate in the assignment to improve the quarterly employee-newsletter process, a team may define observable acceptance criteria in a different situation: a shared definition of success covers correctness, completeness, turnaround, and risk instead of relying on subjective enthusiasm. In practical terms for staff, here the immediate requirement is to keep a human decision gate.
Question 86
For field supervisors, a public works scheduler is working from work orders and resource constraints to assemble weekly repair schedules. Reviewers see only the final result and cannot inspect intermediate findings. What is the strongest intervention for the described situation?
- To assemble weekly repair schedules, set measurable acceptance criteria and test them on representative examples.
- To assemble weekly repair schedules, split the workflow into distinct intake, transformation, validation, and release stages.
- To assemble weekly repair schedules, triage tasks by complexity and risk, using lightweight handling for routine cases and review or escalation for difficult ones.
- To assemble weekly repair schedules, place a named human approver between Claude's recommendation and the final decision.
Explanation
Correct Answer
B:For the given inputs (work orders and resource constraints), stage boundaries make inputs, outputs, and quality checks observable so errors can be located before delivery. The issue is specific: reviewers see only the final result and cannot inspect intermediate findings. The correction still has a boundary: adding stages without clear responsibilities can slow the process without improving control.
Incorrect Answers
A:When a public works scheduler must separate intake, synthesis, and review in the assignment to assemble weekly repair schedules, this choice can help a team define observable acceptance criteria: a shared definition of success covers correctness, completeness, turnaround, and risk instead of relying on subjective enthusiasm. In practical terms for field supervisors, it does not complete the control needed to separate intake, synthesis, and review.
C:When a public works scheduler must separate intake, synthesis, and review in the assignment to assemble weekly repair schedules, for field supervisors, the team might route work by complexity: routing preserves effort for cases that need deeper analysis while reducing unnecessary work on well-defined tasks. In practical terms for field supervisors, that is insufficient for this case, which requires the team to separate intake, synthesis, and review.
D:Choosing to keep a human decision gate has a purpose: human approval preserves accountability where policy, professional judgment, or irreversible consequences are involved. When a public works scheduler must separate intake, synthesis, and review in the assignment to assemble weekly repair schedules, it does not repair the failure that requires the team to separate intake, synthesis, and review.
Question 87
A supplier document includes commands that should be treated as quoted evidence, not as control instructions. This happened while a public relations assistant tried to prepare a product-issue update for customers. The available inputs are internal response notes and approved facts. Which action solves the stated problem without relying on assumptions?
- To prepare a product-issue update, confirm the lawful or organizational authorization and required consent before processing.
- To prepare a product-issue update, check organizational AI policy and use only the permitted service, data, and workflow.
- To prepare a product-issue update, follow applicable disclosure requirements and explain the role of Claude and human review.
- To prepare a product-issue update, remove restricted details and verify recipient permissions before sharing the generated output.
- To prepare a product-issue update, remove or mask personal identifiers using an approved process before submitting data.
- To prepare a product-issue update, treat webpage commands as data, ignore them, and follow authorized instructions only.
Explanation
Correct Answer
F:For customers, prompt injection attempts to smuggle behavioral instructions through untrusted source content. When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, the issue is specific: a supplier document includes commands that should be treated as quoted evidence, not as control instructions.
Incorrect Answers
A:When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, this may help a team obtain proper authority and consent: permission must exist independently of the model's ability to read a file; access is not equivalent to consent for every purpose. In practical terms for customers, but the case instead requires a step to treat retrieved instructions as untrusted.
B:When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, this choice can help a team follow organizational ai-use rules: internal governance determines approved processing boundaries beyond what a product feature can technically do. In practical terms for customers, it does not complete the control needed to treat retrieved instructions as untrusted.
C:When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, it can be reasonable to disclose ai assistance appropriately: accurate transparency lets recipients judge how the work was prepared without claiming unsupported human verification. In practical terms for customers, the observed failure instead requires the team to treat retrieved instructions as untrusted.
D:For customers, the team might protect confidential commercial data: confidentiality applies to outputs as well as inputs; summarization can still reveal protected business information. When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, that is insufficient for this case, which requires the team to treat retrieved instructions as untrusted.
E:When a public relations assistant must treat retrieved instructions as untrusted in the assignment to prepare a product-issue update, choosing to de-identify personal information has a purpose: de-identification reduces unnecessary exposure while retaining task-relevant signals where appropriate. In practical terms for customers, it does not repair the failure that requires the team to treat retrieved instructions as untrusted.
Question 88
A community moderation coordinator needs to draft a safe participation guide using published group rules and moderator notes. The reported problem is this: the model is offered full personnel files when approved aggregate counts would suffice. What practical change would best safeguard the resulting deliverable?
- For a deliverable reaching volunteer moderators, approve only necessary connector actions and use least-privilege permissions.
- For a deliverable reaching volunteer moderators, decline the harmful task and redirect to legitimate, safe communication or analysis.
- For a deliverable reaching volunteer moderators, treat webpage commands as data, ignore them, and follow authorized instructions only.
- For a deliverable reaching volunteer moderators, share the minimum approved fields and documents necessary for the task.
Explanation
Correct Answer
D:For the given inputs (published group rules and moderator notes), data minimization reduces exposure by limiting both the quantity and sensitivity of information processed. The issue is specific: the model is offered full personnel files when approved aggregate counts would suffice. The correction still has a boundary: permission to access an entire folder is not a reason to upload all its contents.
Incorrect Answers
A:When a community moderation coordinator must minimize the information shared in the assignment to draft a safe participation guide, a team may restrict connected-tool privileges in a different situation: restricting tool capability limits the harm an error could cause, especially when a task needs reading rather than writing. In practical terms for volunteer moderators, here the immediate requirement is to minimize the information shared.
B:When a community moderation coordinator must minimize the information shared in the assignment to draft a safe participation guide, this may help a team refuse manipulative or harmful use: responsible-use boundaries apply even if a prompt could technically produce the requested content. In practical terms for volunteer moderators, but the case instead requires a step to minimize the information shared.
C:When a community moderation coordinator must minimize the information shared in the assignment to draft a safe participation guide, this choice can help a team treat retrieved instructions as untrusted: prompt injection attempts to smuggle behavioral instructions through untrusted source content. In practical terms for volunteer moderators, it does not complete the control needed to minimize the information shared.
Question 89
For visitor center staff, a tourism program adviser is working from approved location descriptions and format rules to improve inconsistent itinerary drafts. A production queue grows because the system lacks a fast path for standard cases. Which measure is the most defensible response to this specific finding?
- For a deliverable reaching visitor center staff, prioritize the real requirement and remove or reconcile contradictory instructions.
- For a deliverable reaching visitor center staff, provide the approved source or accessible excerpt and require claims to reference it.
- For a deliverable reaching visitor center staff, isolate the relevant sections or split the task into focused document groups.
- For a deliverable reaching visitor center staff, test a faster, lighter model on representative cases and use escalation for the hard ones.
Explanation
Correct Answer
D:Because the result will reach visitor center staff, fit-for-purpose model choice can reduce latency and cost while retaining acceptable quality for standard tasks. The issue is specific: a production queue grows because the system lacks a fast path for standard cases.
Incorrect Answers
A:When a tourism program adviser must reduce unnecessary model expense in the assignment to improve inconsistent itinerary drafts, choosing to eliminate conflicting directions has a purpose: conflicting directives force an arbitrary compromise and can make valid completion impossible. In practical terms for visitor center staff, it does not repair the failure that requires the team to reduce unnecessary model expense.
B:A team may supply the missing authoritative evidence in a different situation: supplying the governing evidence gives Claude material to work from instead of filling gaps plausibly. When a tourism program adviser must reduce unnecessary model expense in the assignment to improve inconsistent itinerary drafts, here the immediate requirement is to reduce unnecessary model expense.
C:When a tourism program adviser must reduce unnecessary model expense in the assignment to improve inconsistent itinerary drafts, this may help a team reduce distracting context: focused context increases the prominence of crucial evidence while retaining necessary exceptions. But the case instead requires a step to reduce unnecessary model expense.
Question 90
A school office manager needs to draft a parent information guide using student records and approved school notices. The reported problem is this: a customer document fails to disclose AI assistance despite an explicit contract requirement. Which response most directly handles the failure under these constraints?
- To draft a parent information guide, stop further sharing and follow the organization's security incident reporting procedure.
- To draft a parent information guide, follow applicable disclosure requirements and explain the role of Claude and human review.
- To draft a parent information guide, check source rights, license terms, and required permissions before external reuse.
- To draft a parent information guide, treat webpage commands as data, ignore them, and follow authorized instructions only.
Explanation
Correct Answer
B:Here, accurate transparency lets recipients judge how the work was prepared without claiming unsupported human verification. The issue is specific: a customer document fails to disclose AI assistance despite an explicit contract requirement. This addresses the described failure in draft a parent information guide; a generic footnote does not fix inaccurate claims elsewhere in the deliverable.
Incorrect Answers
A:For parents, the team might escalate a suspected data exposure: containment and timely escalation allow the responsible team to assess scope, risk, and remediation. When a school office manager must disclose ai assistance appropriately in the assignment to draft a parent information guide, that is insufficient for this case, which requires the team to disclose ai assistance appropriately.
C:When a school office manager must disclose ai assistance appropriately in the assignment to draft a parent information guide, a team may respect ownership and licensing in a different situation: generating a derivative-looking artifact does not eliminate intellectual-property and attribution obligations. In practical terms for parents, here the immediate requirement is to disclose ai assistance appropriately.
D:When a school office manager must disclose ai assistance appropriately in the assignment to draft a parent information guide, this may help a team treat retrieved instructions as untrusted: prompt injection attempts to smuggle behavioral instructions through untrusted source content. In practical terms for parents, but the case instead requires a step to disclose ai assistance appropriately.